Every once in awhile, I see an IVR developer on the forum mention the possibility of building an alphanumeric IVR grammar that will allow users to spell out letters and numbers over the phone.
My response to that is this: Have you ever tried to spell out information to someone over the phone?
For example:
Person A: Hi! Welcome to the Bank of Money. How can I help you today?
Person B: Hi there. I’m having a problem with my account.
Person A: Sure. I can help you with that. Can I have you spell out your first name?
Person B: Valerie. V-a-l-e-r-i-e.
Person A: I’m sorry, I missed some of that. Was it V as in Victor? Or B as in Banana?
Person B: V as in Victor.
Person A: Last name?
Person B: Munn. M-u-n-n.
Person A: M as in Mary. U as in Umbrella. M as in Mary. M as in Mary?
Person B: No. M as in Mary. U as in Umbrella. N as in Nancy. N as in Nancy.
Person A: Ah ok. Sorry for the misunderstanding, ma’am.
…
…
…
You get the point. The problem is that even humans have trouble understanding spelling from other humans over the phone as there can be constant mix-ups between m’s and n’s, b’s and v’s, and f’s and s’s. Also, add to that the possibility of a person having a thick accent while spelling out words.
So, if humans have difficulty spelling out words, you can expect that an IVR speech recognition engine would have a problem capturing a spelling grammar. One possible solution is to implement an IVR grammar that uses the military phonetic alphabet to try to capture these spelling grammars.
For example, some IVR code to demonstrate:
<rule id=”alpha” scope=”private”>
<one-of>
<item>A<tag>SWI_literal=”A”</tag></item>
<item>A as in alpha<tag>SWI_literal=”A”</tag></item>
</one-of>
</rule>
However, even this solution would be difficult as an IVR developer shouldn’t expect general users to know of the military alphabet.
Leave a Reply